Scientific Profiling based on Semantic Analysis in Social Networks

Laurens De Vocht

SupervisorsGonzalo ParraSelver Softic

PromotorsErik Duval

Martin Ebner

April 11, 2011

‣Problem Statement‣Solution‣Architecture‣Implementation‣Evaluation‣Conclusion‣Discussion

2

Agenda

Problem Statement

3

Who has scientific information relevant for me?

Web users generate a massive unstructured data flow

?

Solution

4

Community approvedontologies: FOAF, SIOC

Annotate Data from Social Networks

Linked Open Data

Scientific Profiling Framework

Applications

Connect People and Resources that share Scientific Affinities

5

Architecture

‣Overview‣Grabeeter‣Framework‣API

Architecture: Overview

6

Linked Open Data Cloud

Social Networks

Aggregate

Archived/Cached Data Linked Data

Annotate

Interlink

Scientific InformationAnalyse

Output Format

PublishFramework

Architecture: Overview

6

DBPediaColinda

GeoNamesTwitter

Aggregate

Grabeeter Semantic Profiling NetworkAnnotate

Interlink

Scientific Profiling APIAnalyse

JSONRDF (XML)

Publish

Linked Open Data Cloud

Social Networks

Aggregate

Archived/Cached Data Linked Data

Annotate

Interlink

Scientific InformationAnalyse

Output Format

PublishFramework

Architecture: Overview

7

DBPediaColinda

GeoNamesTwitter

Aggregate

Grabeeter Semantic Profiling NetworkAnnotate

Interlink

Scientific Profiling APIAnalyse

JSONRDF (XML)

Publish

Linked Open Data Cloud

Social Networks

Aggregate

Archived/Cached Data Linked Data

Annotate

Interlink

Scientific InformationAnalyse

Output Format

PublishFramework

Architecture: Grabeeter

8

= Twitter aggregation & archiving tool(developed at TUGraz)

Architecture: Grabeeter

8

= Twitter aggregation & archiving tool(developed at TUGraz)

Architecture: Overview

9

DBPediaColinda

GeoNamesTwitter

Aggregate

Grabeeter Semantic Profiling NetworkAnnotate

Interlink

Scientific Profiling APIAnalyse

JSONRDF (XML)

Publish

Linked Open Data Cloud

Social Networks

Aggregate

Archived/Cached Data Linked Data

Annotate

Interlink

Scientific InformationAnalyse

Output Format

PublishFramework

Architecture: Framework

10

Interlinking

Analysis

Extraction

Grabeeter

Applications

Programming Interface

SQL Queries Triplification SPARQL Queries

RDF Store

High Level Queries

Architecture: API

11

‣ User Profile

http://api.semanticprofiling.net/profile.php?user=screen_name

‣ Discover people, events...

http://api.semanticprofiling.net/discovery.php?

find=persons|events|popular_friends|popular_mentions|popular_events

user=screen_name

‣ Register new Twitter user

http://api.semanticprofiling.net/register.php?user=twitter_user

‣ Event Details

http://api.semanticprofiling.net/event.php?name=event_name

Architecture: API

12

Architecture: API

13

14

Implementation‣Hashtags as Identifiers‣not always strong or consistent enough‣properties of good hashtags formalized‣helpful in assessment of valuable identifiers

‣Expert Search/Profiling with Linked Data‣aggregate and analyze certain types of data‣need to surpass limits of closed data sets‣LOD delivers multi-purpose data

ReferencesLaniado, D., Mika. P.: Making sense of twitter. In: The Semantic Web–ISWC 2010 (2010)Stankovic M., Wagner C., Laublet P., Jovanovic J.: Looking for experts? what can linked data do for you. In: Linked Data on the Web.

15

Evaluation

‣Approach‣Usability Case Study‣Search Quality‣Results so far

16

Evaluation: Approach

‣Test usability & search quality‣Web application: “Scientific Affinity Browser” as case study.‣Development in 3 phases:‣1. User Interface only‣2. Fixed Data Set‣3. Integrated with Framework

Evaluation: Usability Case Study

17

Connecting researchers based on shared scientific events (conferences)

Researchers

Scientific Profiling

Researcher (User)

User Model Event ModelScientific Conferences Resource

Profiler/Analyzer

18

Evaluation: Search quality

‣Satisfaction questionnaireUsers rate their search results‣Measure Relevance‣Recall‣Precision

Evaluation: Results so far

19

‣Definitely useful application‣Use of the map view makes sense‣People - Event split confusing‣View of own profile‣not a suitable starting point‣only useful in comparison‣shouldn’t be always visible‣Person-specific affinities‣too much hidden

Evaluation: Results so far

20

Challenges

21

‣Rank tags

by importance, not just frequency of use

‣Visualization

of similarities and links between users and entities

‣Detect communities

using a graph-based algorithm on the linked data network

Schedule

22

Iteration 8 feb 21 - mar 2Implement

framework feedback

Iteration 9 mar 3 - mar 16 Literature study 2

Iteration 10 mar 17 - apr 8Evaluation phase 1 & Report/presentation

Iteration 11 apr 9 - apr 28 Evaluation phase 2

Iteration 12 apr 29 - may 15 Evaluation phase 3

Conclusion

23

‣ Framework designed to support a broad range of

social scientific semantic-based applications

‣Realized with current state of the art technologies

‣ Interlinking with Linked Open Data Cloud

‣ Evaluation case study: scientific affinity browser web app

‣ Follow project blog http://blog.semanticprofiling.net

‣All References: see reports